On 02/11/2018 07:41 PM, John Larkin wrote:> On Sun, 11 Feb 2018 17:33:30 -0500, Phil Hobbs > <pcdhSpamMeSenseless@electrooptical.net> wrote: > >> On 02/11/2018 04:13 PM, John Larkin wrote: >>> On Sun, 11 Feb 2018 15:44:14 -0500, Phil Hobbs >>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>> >>>> On 02/11/2018 02:58 PM, John Larkin wrote: >>>>> On Sun, 11 Feb 2018 13:17:25 -0500, Phil Hobbs >>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>>>> >>>>>> On 02/11/2018 12:52 PM, John Larkin wrote: >>>>>>> On Sun, 11 Feb 2018 12:32:55 -0500, Phil Hobbs >>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>>>>>> >>>>>>>> On 02/11/2018 11:34 AM, John Larkin wrote: >>>>>>>>> On Sun, 11 Feb 2018 08:05:11 -0500, Phil Hobbs >>>>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>>>>>>>> >>>>>>>>>> On 02/10/2018 08:36 PM, John Larkin wrote: >>>>>>>>>>> On Sat, 10 Feb 2018 19:39:32 -0500, Phil Hobbs >>>>>>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>>>>>>>>>> >>>>>>>>>>>> On 02/10/18 18:41, John Larkin wrote: >>>>>>>>>>>>> On Sat, 10 Feb 2018 17:18:55 -0500, Phil Hobbs >>>>>>>>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>>>>>>>>>>>> >>>>>>>>>>>>>> Is horrible. >>>>>>>>>>>>>> >>>>>>>>>>>>>> I'm just debugging a nice diode laser controller for one >>>>>>>>>>>>>> customer and getting ready to do another one for another (much >>>>>>>>>>>>>> better) customer. It uses a ZXTP25020 PNP with a biggish >>>>>>>>>>>>>> degeneration resistor and two-pole bypassing of the base to get >>>>>>>>>>>>>> low noise at high frequency. >>>>>>>>>>>>>> >>>>>>>>>>>>>> Turns out the 1/f noise is atrocious. The total noise from 2 >>>>>>>>>>>>>> Hz on up is only a couple of PPM, but it should be a factor of >>>>>>>>>>>>>> 10 better than that. With a total base resistance of 1k, the >>>>>>>>>>>>>> 1/f corner at 40 mA I_C is over 10 kHz. >>>>>>>>>>>>>> >>>>>>>>>>>>>> The bias network looks like this: >>>>>>>>>>>>>> >>>>>>>>>>>>>> +9V 0----------*--------*-------* | | | | | >>>>>>>>>>>>>> | | | R | 4.7 | 3.3 R 39 ohm TF | nF | >>>>>>>>>>>>>> nF R CCC CCC | CCC CCC | | | >>>>>>>>>>>>>> | / | | |< ZXTP25020 From 0--RRRR--*--RRRR--*---- >>>>>>>>>>>>>> | Op Amp |\ 499 499 | \ | >>>>>>>>>>>>>> | --- \ / --> V --> ----- | | GND >>>>>>>>>>>>>> >>>>>>>>>>>>>> The emitter resistor can drop up to about 5V. >>>>>>>>>>>>>> >>>>>>>>>>>>>> I've got some higher-beta transistors on order (FZT788B), and >>>>>>>>>>>>>> will reduce the impedance of the bias network, but I was fairly >>>>>>>>>>>>>> shocked that the 1/f noise was so bad. >>>>>>>>>>>>>> >>>>>>>>>>>>>> Cheers >>>>>>>>>>>>>> >>>>>>>>>>>>>> Phil Hobbs >>>>>>>>>>>>> >>>>>>>>>>>>> That's a pretty fast transistor; could it be oscillating? >>>>>>>>>>>> >>>>>>>>>>>> Don't think so. I have a 1 GHz scope across the laser, and didn't >>>>>>>>>>>> see anything like that, and have applied thermal-digital >>>>>>>>>>>> analysis(*) to the whole thing. I'll try going to 49.9 ohms and 47 >>>>>>>>>>>> nF and see if it gets better, which it should. I might need to use >>>>>>>>>>>> a TCA0372 booster to drive 5-ohm resistors. :( >>>>>>>>>>>> >>>>>>>>>>>> Cheers >>>>>>>>>>>> >>>>>>>>>>>> Phil Hobbs >>>>>>>>>>>> >>>>>>>>>>>> (*) i.e. heat gun, cold spray, and fingers >>>>>>>>>>> >>>>>>>>>>> Yeah, the base current noise into the 1K resistor string could make >>>>>>>>>>> a lot of noise... assuming the signal from the opamp is perfect. >>>>>>>>>>> >>>>>>>>>>> Is the overall feedback optical? >>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>> >>>>>>>>>> Eventually, but at the moment it's coming from the emitter resistor, so >>>>>>>>>> that the loop doesn't null out the base current noise even within its >>>>>>>>>> bandwidth. >>>>>>>>>> >>>>>>>>>> The idea is to make the bias current way sub-Poissonian, which if done >>>>>>>>>> right results in useful amounts of amplitude squeezing of the light, >>>>>>>>>> i.e. your detected photocurrent can actually be a few dB quieter than >>>>>>>>>> full shot noise if you keep the optical loss low enough. It's more of a >>>>>>>>>> checklist feature for most uses, of course, but as you've pointed out >>>>>>>>>> there are a lot of crappy diode laser controllers out there, selling for >>>>>>>>>> cheap, so building one that just makes the problems magically go away is >>>>>>>>>> attractive. >>>>>>>>>> >>>>>>>>>> Building a new-to-me kind of high dynamic range system always seems to >>>>>>>>>> produce a bit of buried treasure at first. The ZXTP25020 has super-low >>>>>>>>>> Rbb' and Ree', so it would be an excellent candidate except for the 1/f >>>>>>>>>> noise. If the higher-beta device and lower drive impedance doesn't >>>>>>>>>> improve it enough, I'll have to try Darlingtonizing it with one of those >>>>>>>>>> nice SiGe:C devices to return the base current to the collector circuit. >>>>>>>>>> >>>>>>>>>> That'd need a board spin though, and I'm super busy at the moment. >>>>>>>>>> >>>>>>>>>> Cheers >>>>>>>>>> >>>>>>>>>> Phil Hobbs >>>>>>>>> >>>>>>>>> How about a Sziklai pair sort of thing, with a BF862 and a PNP? Or a >>>>>>>>> Pfet-PNP darlington? Disappear the base current. >>>>>>>> >>>>>>>> I need it to be a PNP-type when it's done, because the lasers are >>>>>>>> usually grounded-cathode. (Keeping the high frequency output impedance >>>>>>>> high is the name of the squeezing game--with an emitter follower you get >>>>>>>> twice the shot noise.) If I could get good PFETs, that would be great, >>>>>>>> but they're scarce these days and their noise tends to be horrible. >>>>>>>> >>>>>>>> The Darlington-bandaid approach could be done with an NPN and a couple >>>>>>>> of auxiliary current sources, but it would be a bit of a mess. The >>>>>>>> advantage would be that (done right) it would have the noise of the >>>>>>>> BFP640, which is very low. >>>>>>>> >>>>>>>> Tomorrow I'll try reducing the base impedance by 10x and see if that >>>>>>>> fixes it. Because of the 39-ohm emitter resistor, it doesn't make sense >>>>>>>> to go very much lower than that. >>>>>>>> >>>>>>>> One nice thing about 1/f noise is that a 10x amplitude improvement gets >>>>>>>> you 100x lower corner frequency. At that point I can put in a >>>>>>>> low-frequency bandaid if I need to, such as sensing the voltage across >>>>>>>> the base resistor and dumping the equivalent into the emitter so that >>>>>>>> the op amp tracks out the LF noise. The trick will be to avoid a noise >>>>>>>> peak at the crossover frequency of the bandaid. >>>>>>>> >>>>>>>> Cheers >>>>>>>> >>>>>>>> Phil Hobbs >>>>>>> >>>>>>> Here's an idea that I've never actually built: >>>>>>> >>>>>>> https://www.dropbox.com/s/lr3aekzziecz7pz/Isrc_cascode_Ib.JPG?raw=1 >>>>>>> >>>>>>> I don't remember what those two caps are for. >>>>>> >>>>>> >>>>>> Interesting. It has the current noise of the FET at low frequency, >>>>>> though, and looks like it has full shot noise at high frequency on >>>>>> account of the caps. >>>>>> >>>>>> I'd be looking at something vaguely of that kind for the bandaid >>>>>> approach--next time I have a Saturday afternoon free I'll have a whack >>>>>> at it. >>>>>> >>>>>> Cheers >>>>>> >>>>>> Phil Hobbs >>>>> >>>>> Have you considered a bootstrap current source? That solves a lot of >>>>> problems. Adding an inductor somewhere makes it better at high >>>>> frequencies, so you can concentrate on the low end. Any copper TC >>>>> error is easily compensated. >>>> >>>> The bootstrap sources I know about use op amps to control things at all >>>> frequencies, so they're at least 10-15 dB noisier at medium-to-high >>>> frequency than a transistor with Rbb' of an ohm. It's really just the >>>> 1/f noise that's the problem. An inductor is a possibility, though--it >>>> might very well help up around f_T. >>>> >>>> Cheers >>>> >>>> Phil Hobbs >>> >>> Chopamp? >> >> The simple op amp + BJT + local feedback approach has really stellar >> noise performance at high frequency. The issue at low frequency is that >> the base current 1/f noise doesn't get tracked out by the op amp loop >> because it appears in the collector circuit but not in the emitter >> circuit, which is where the FB is applied. >> >> ISTM there are four ways of fixing this wart without degrading the HF >> noise. From easiest to hardest: >> >> 1. Reduce the 1/f noise by reducing the base circuit impedance. >> >> 2. Do some Darlingtonish thing to return the base current to the >> collector circuit without adding significant voltage noise. >> >> 3. Use some more complicated local feedback scheme to synthesize a copy >> of the base current and stick it into the emitter circuit so that the >> main loop tracks it out. >> >> 4. Come up with some completely new scheme. >> >> Because the laser current-tunes, its linewidth is degraded more by >> strong low-frequency components than by the same noise power spread out >> over the full bandwidth. (This is just like an FM transmitter.) Thus >> it would be worth trading off a bit of HF noise to fix this issue, but >> I'm greedy and want to have it all. ;) >> >> Chopamps tend to have low frequency noise around 20-30 nV in 1 Hz all >> the way down to DC, which isn't bad at all--it would flatten out my >> noise below about 20 Hz. My faves are the OPA2188, which is a 32V dual, >> and the OPA378, which is a bit noisier in the flatband but whose noise >> doesn't rise at all at low frequency, at least down to 10 uHz (~1 >> cycle/day) which is as far as I've measured it. >> >> I'd be very interested to hear about yours and others' fave >> chopamps--there's a depressing amount of specsmanship going on in their >> datasheets. >> >> Cheers >> >> Phil Hobbs > > We've use a bunch of them, but they tend to have high wideband noise > and sometimes weird charge injection problems. The well-behaved > LTC2057 is "only" 11 nv/rthz, but the ADI parts are around twice that, > except the ADA4638, which is 66! > > The 2057 has a noise spike around 50 KHz, so for extreme performance > I'd think about compounding it with another amp. Let the chopamp take > over below the 1/f corner of another amp.I've done that before. One issue is that the chopamp spikes are mostly current noise, so you have to bypass the chopamp inputs to keep the spikes out of the rest of the circuit. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC / Hobbs ElectroOptics Optics, Electro-optics, Photonics, Analog Electronics Briarcliff Manor NY 10510 http://electrooptical.net https://hobbs-eo.com
BJT base current 1/f noise
Started by ●February 10, 2018
Reply by ●February 12, 20182018-02-12
Reply by ●February 12, 20182018-02-12
On 02/12/2018 09:09 AM, George Herold wrote:> On Sunday, February 11, 2018 at 5:33:47 PM UTC-5, Phil Hobbs wrote: >> On 02/11/2018 04:13 PM, John Larkin wrote: >>> On Sun, 11 Feb 2018 15:44:14 -0500, Phil Hobbs >>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>> >>>> On 02/11/2018 02:58 PM, John Larkin wrote: >>>>> On Sun, 11 Feb 2018 13:17:25 -0500, Phil Hobbs >>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>>>> >>>>>> On 02/11/2018 12:52 PM, John Larkin wrote: >>>>>>> On Sun, 11 Feb 2018 12:32:55 -0500, Phil Hobbs >>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>>>>>> >>>>>>>> On 02/11/2018 11:34 AM, John Larkin wrote: >>>>>>>>> On Sun, 11 Feb 2018 08:05:11 -0500, Phil Hobbs >>>>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>>>>>>>> >>>>>>>>>> On 02/10/2018 08:36 PM, John Larkin wrote: >>>>>>>>>>> On Sat, 10 Feb 2018 19:39:32 -0500, Phil Hobbs >>>>>>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>>>>>>>>>> >>>>>>>>>>>> On 02/10/18 18:41, John Larkin wrote: >>>>>>>>>>>>> On Sat, 10 Feb 2018 17:18:55 -0500, Phil Hobbs >>>>>>>>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>>>>>>>>>>>> >>>>>>>>>>>>>> Is horrible. >>>>>>>>>>>>>> >>>>>>>>>>>>>> I'm just debugging a nice diode laser controller for one >>>>>>>>>>>>>> customer and getting ready to do another one for another (much >>>>>>>>>>>>>> better) customer. It uses a ZXTP25020 PNP with a biggish >>>>>>>>>>>>>> degeneration resistor and two-pole bypassing of the base to get >>>>>>>>>>>>>> low noise at high frequency. >>>>>>>>>>>>>> >>>>>>>>>>>>>> Turns out the 1/f noise is atrocious. The total noise from 2 >>>>>>>>>>>>>> Hz on up is only a couple of PPM, but it should be a factor of >>>>>>>>>>>>>> 10 better than that. With a total base resistance of 1k, the >>>>>>>>>>>>>> 1/f corner at 40 mA I_C is over 10 kHz. >>>>>>>>>>>>>> >>>>>>>>>>>>>> The bias network looks like this: >>>>>>>>>>>>>> >>>>>>>>>>>>>> +9V 0----------*--------*-------* | | | | | >>>>>>>>>>>>>> | | | R | 4.7 | 3.3 R 39 ohm TF | nF | >>>>>>>>>>>>>> nF R CCC CCC | CCC CCC | | | >>>>>>>>>>>>>> | / | | |< ZXTP25020 From 0--RRRR--*--RRRR--*---- >>>>>>>>>>>>>> | Op Amp |\ 499 499 | \ | >>>>>>>>>>>>>> | --- \ / --> V --> ----- | | GND >>>>>>>>>>>>>> >>>>>>>>>>>>>> The emitter resistor can drop up to about 5V. >>>>>>>>>>>>>> >>>>>>>>>>>>>> I've got some higher-beta transistors on order (FZT788B), and >>>>>>>>>>>>>> will reduce the impedance of the bias network, but I was fairly >>>>>>>>>>>>>> shocked that the 1/f noise was so bad. >>>>>>>>>>>>>> >>>>>>>>>>>>>> Cheers >>>>>>>>>>>>>> >>>>>>>>>>>>>> Phil Hobbs >>>>>>>>>>>>> >>>>>>>>>>>>> That's a pretty fast transistor; could it be oscillating? >>>>>>>>>>>> >>>>>>>>>>>> Don't think so. I have a 1 GHz scope across the laser, and didn't >>>>>>>>>>>> see anything like that, and have applied thermal-digital >>>>>>>>>>>> analysis(*) to the whole thing. I'll try going to 49.9 ohms and 47 >>>>>>>>>>>> nF and see if it gets better, which it should. I might need to use >>>>>>>>>>>> a TCA0372 booster to drive 5-ohm resistors. :( >>>>>>>>>>>> >>>>>>>>>>>> Cheers >>>>>>>>>>>> >>>>>>>>>>>> Phil Hobbs >>>>>>>>>>>> >>>>>>>>>>>> (*) i.e. heat gun, cold spray, and fingers >>>>>>>>>>> >>>>>>>>>>> Yeah, the base current noise into the 1K resistor string could make >>>>>>>>>>> a lot of noise... assuming the signal from the opamp is perfect. >>>>>>>>>>> >>>>>>>>>>> Is the overall feedback optical? >>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>> >>>>>>>>>> Eventually, but at the moment it's coming from the emitter resistor, so >>>>>>>>>> that the loop doesn't null out the base current noise even within its >>>>>>>>>> bandwidth. >>>>>>>>>> >>>>>>>>>> The idea is to make the bias current way sub-Poissonian, which if done >>>>>>>>>> right results in useful amounts of amplitude squeezing of the light, >>>>>>>>>> i.e. your detected photocurrent can actually be a few dB quieter than >>>>>>>>>> full shot noise if you keep the optical loss low enough. It's more of a >>>>>>>>>> checklist feature for most uses, of course, but as you've pointed out >>>>>>>>>> there are a lot of crappy diode laser controllers out there, selling for >>>>>>>>>> cheap, so building one that just makes the problems magically go away is >>>>>>>>>> attractive. >>>>>>>>>> >>>>>>>>>> Building a new-to-me kind of high dynamic range system always seems to >>>>>>>>>> produce a bit of buried treasure at first. The ZXTP25020 has super-low >>>>>>>>>> Rbb' and Ree', so it would be an excellent candidate except for the 1/f >>>>>>>>>> noise. If the higher-beta device and lower drive impedance doesn't >>>>>>>>>> improve it enough, I'll have to try Darlingtonizing it with one of those >>>>>>>>>> nice SiGe:C devices to return the base current to the collector circuit. >>>>>>>>>> >>>>>>>>>> That'd need a board spin though, and I'm super busy at the moment. >>>>>>>>>> >>>>>>>>>> Cheers >>>>>>>>>> >>>>>>>>>> Phil Hobbs >>>>>>>>> >>>>>>>>> How about a Sziklai pair sort of thing, with a BF862 and a PNP? Or a >>>>>>>>> Pfet-PNP darlington? Disappear the base current. >>>>>>>> >>>>>>>> I need it to be a PNP-type when it's done, because the lasers are >>>>>>>> usually grounded-cathode. (Keeping the high frequency output impedance >>>>>>>> high is the name of the squeezing game--with an emitter follower you get >>>>>>>> twice the shot noise.) If I could get good PFETs, that would be great, >>>>>>>> but they're scarce these days and their noise tends to be horrible. >>>>>>>> >>>>>>>> The Darlington-bandaid approach could be done with an NPN and a couple >>>>>>>> of auxiliary current sources, but it would be a bit of a mess. The >>>>>>>> advantage would be that (done right) it would have the noise of the >>>>>>>> BFP640, which is very low. >>>>>>>> >>>>>>>> Tomorrow I'll try reducing the base impedance by 10x and see if that >>>>>>>> fixes it. Because of the 39-ohm emitter resistor, it doesn't make sense >>>>>>>> to go very much lower than that. >>>>>>>> >>>>>>>> One nice thing about 1/f noise is that a 10x amplitude improvement gets >>>>>>>> you 100x lower corner frequency. At that point I can put in a >>>>>>>> low-frequency bandaid if I need to, such as sensing the voltage across >>>>>>>> the base resistor and dumping the equivalent into the emitter so that >>>>>>>> the op amp tracks out the LF noise. The trick will be to avoid a noise >>>>>>>> peak at the crossover frequency of the bandaid. >>>>>>>> >>>>>>>> Cheers >>>>>>>> >>>>>>>> Phil Hobbs >>>>>>> >>>>>>> Here's an idea that I've never actually built: >>>>>>> >>>>>>> https://www.dropbox.com/s/lr3aekzziecz7pz/Isrc_cascode_Ib.JPG?raw=1 >>>>>>> >>>>>>> I don't remember what those two caps are for. >>>>>> >>>>>> >>>>>> Interesting. It has the current noise of the FET at low frequency, >>>>>> though, and looks like it has full shot noise at high frequency on >>>>>> account of the caps. >>>>>> >>>>>> I'd be looking at something vaguely of that kind for the bandaid >>>>>> approach--next time I have a Saturday afternoon free I'll have a whack >>>>>> at it. >>>>>> >>>>>> Cheers >>>>>> >>>>>> Phil Hobbs >>>>> >>>>> Have you considered a bootstrap current source? That solves a lot of >>>>> problems. Adding an inductor somewhere makes it better at high >>>>> frequencies, so you can concentrate on the low end. Any copper TC >>>>> error is easily compensated. >>>> >>>> The bootstrap sources I know about use op amps to control things at all >>>> frequencies, so they're at least 10-15 dB noisier at medium-to-high >>>> frequency than a transistor with Rbb' of an ohm. It's really just the >>>> 1/f noise that's the problem. An inductor is a possibility, though--it >>>> might very well help up around f_T. >>>> >>>> Cheers >>>> >>>> Phil Hobbs >>> >>> Chopamp? >> > >> The simple op amp + BJT + local feedback approach has really stellar >> noise performance at high frequency. The issue at low frequency is that >> the base current 1/f noise doesn't get tracked out by the op amp loop >> because it appears in the collector circuit but not in the emitter >> circuit, which is where the FB is applied. > > Phil, the above bothered me. Isn't the emitter current the sum of > base and collector? But I think I understand... base current 1/f > noise appears in emitter, and gets taken care of by FB... but that > screws up the collector current which has no 1/f noise. > Or is my head on backwards this morning?You can look at it either way. Inside the op amp loop's control bandwidth, the voltage across the emitter resistor is held still by overall feedback, so regardless of the noise voltage drop across the base resistor, all the 1/f noise has to come out the collector. Outside that bandwidth, there's a narrow region where the base resistors contribute, but then the bypass caps take over and hold the voltage across the emitter resistance constant again. So basically the noise comes out the collector at all frequencies as long as the emitter resistor is dropping several times kT/e.>> >> ISTM there are four ways of fixing this wart without degrading the HF >> noise. From easiest to hardest: >> >> 1. Reduce the 1/f noise by reducing the base circuit impedance.Didn't help, for reasons given above.>> >> 2. Do some Darlingtonish thing to return the base current to the >> collector circuit without adding significant voltage noise.Using an MPS4250 Darlington helped a lot with the 1/f noise, but the flatband noise is now only about 6 dB below shot noise, which is a pity. Good enough for this rev, though. Now I just need to find a suitable Darlington in SOT89 or SOT223. Next time round I'll use another of the power devices as the input transistor, and add some extra base current to it, but not enough for its 1/f noise to dominate.>> >> 3. Use some more complicated local feedback scheme to synthesize a copy >> of the base current and stick it into the emitter circuit so that the >> main loop tracks it out. >> >> 4. Come up with some completely new scheme. > > You could feed back with a sense resistor on the same side as the load. > I guess that means either floating the load (laser diode.. not good) > or floating the sense R. (or I'm confused) > > George H.-- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC / Hobbs ElectroOptics Optics, Electro-optics, Photonics, Analog Electronics Briarcliff Manor NY 10510 http://electrooptical.net https://hobbs-eo.com
Reply by ●February 12, 20182018-02-12
On 02/12/2018 02:23 AM, piglet wrote:> On 10/02/2018 22:18, Phil Hobbs wrote: >> Is horrible. >> >> I'm just debugging a nice diode laser controller for one customer and >> getting ready to do another one for another (much better) customer. It >> uses a ZXTP25020 PNP with a biggish degeneration resistor and two-pole >> bypassing of the base to get low noise at high frequency. >> >> Turns out the 1/f noise is atrocious. The total noise from 2 Hz on up >> is only a couple of PPM, but it should be a factor of 10 better than >> that. With a total base resistance of 1k, the 1/f corner at 40 mA I_C >> is over 10 kHz. >> >> The bias network looks like this: >> >> +9V 0----------*--------*-------* >> | | | >> | | | >> | | R >> | 4.7 | 3.3 R 39 ohm TF >> | nF | nF R >> CCC CCC | >> CCC CCC | >> | | | / >> | | |< ZXTP25020 >> From 0--RRRR--*--RRRR--*---- | >> Op Amp |\ >> 499 499 | \ >> | >> | >> --- >> \ / --> >> V --> >> ----- >> | >> | >> GND >> >> The emitter resistor can drop up to about 5V. >> >> I've got some higher-beta transistors on order (FZT788B), and will >> reduce the impedance of the bias network, but I was fairly shocked that >> the 1/f noise was so bad. >> >> Cheers >> >> Phil Hobbs >> > > Would paralleling a few help? Decorrelated noise and all that? > > piglet >I don't know. If the 1/f noise is an areal effect, chopping the die up into smaller ones wouldn't make much difference, other things being equal. If it's a perimeter effect, parallelling devices would make it worse. It would need to be some sort of collective effect for a single die to be worse. I can't think of a mechanism for that offhand, but that doesn't mean it's impossible. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC / Hobbs ElectroOptics Optics, Electro-optics, Photonics, Analog Electronics Briarcliff Manor NY 10510 http://electrooptical.net https://hobbs-eo.com
Reply by ●February 12, 20182018-02-12
On Mon, 12 Feb 2018 11:07:57 -0500, Phil Hobbs <pcdhSpamMeSenseless@electrooptical.net> wrote:>On 02/11/2018 07:41 PM, John Larkin wrote: >> On Sun, 11 Feb 2018 17:33:30 -0500, Phil Hobbs >> <pcdhSpamMeSenseless@electrooptical.net> wrote: >> >>> On 02/11/2018 04:13 PM, John Larkin wrote: >>>> On Sun, 11 Feb 2018 15:44:14 -0500, Phil Hobbs >>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>>> >>>>> On 02/11/2018 02:58 PM, John Larkin wrote: >>>>>> On Sun, 11 Feb 2018 13:17:25 -0500, Phil Hobbs >>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>>>>> >>>>>>> On 02/11/2018 12:52 PM, John Larkin wrote: >>>>>>>> On Sun, 11 Feb 2018 12:32:55 -0500, Phil Hobbs >>>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>>>>>>> >>>>>>>>> On 02/11/2018 11:34 AM, John Larkin wrote: >>>>>>>>>> On Sun, 11 Feb 2018 08:05:11 -0500, Phil Hobbs >>>>>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>>>>>>>>> >>>>>>>>>>> On 02/10/2018 08:36 PM, John Larkin wrote: >>>>>>>>>>>> On Sat, 10 Feb 2018 19:39:32 -0500, Phil Hobbs >>>>>>>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>>>>>>>>>>> >>>>>>>>>>>>> On 02/10/18 18:41, John Larkin wrote: >>>>>>>>>>>>>> On Sat, 10 Feb 2018 17:18:55 -0500, Phil Hobbs >>>>>>>>>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>>>>>>>>>>>>> >>>>>>>>>>>>>>> Is horrible. >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> I'm just debugging a nice diode laser controller for one >>>>>>>>>>>>>>> customer and getting ready to do another one for another (much >>>>>>>>>>>>>>> better) customer. It uses a ZXTP25020 PNP with a biggish >>>>>>>>>>>>>>> degeneration resistor and two-pole bypassing of the base to get >>>>>>>>>>>>>>> low noise at high frequency. >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> Turns out the 1/f noise is atrocious. The total noise from 2 >>>>>>>>>>>>>>> Hz on up is only a couple of PPM, but it should be a factor of >>>>>>>>>>>>>>> 10 better than that. With a total base resistance of 1k, the >>>>>>>>>>>>>>> 1/f corner at 40 mA I_C is over 10 kHz. >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> The bias network looks like this: >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> +9V 0----------*--------*-------* | | | | | >>>>>>>>>>>>>>> | | | R | 4.7 | 3.3 R 39 ohm TF | nF | >>>>>>>>>>>>>>> nF R CCC CCC | CCC CCC | | | >>>>>>>>>>>>>>> | / | | |< ZXTP25020 From 0--RRRR--*--RRRR--*---- >>>>>>>>>>>>>>> | Op Amp |\ 499 499 | \ | >>>>>>>>>>>>>>> | --- \ / --> V --> ----- | | GND >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> The emitter resistor can drop up to about 5V. >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> I've got some higher-beta transistors on order (FZT788B), and >>>>>>>>>>>>>>> will reduce the impedance of the bias network, but I was fairly >>>>>>>>>>>>>>> shocked that the 1/f noise was so bad. >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> Cheers >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> Phil Hobbs >>>>>>>>>>>>>> >>>>>>>>>>>>>> That's a pretty fast transistor; could it be oscillating? >>>>>>>>>>>>> >>>>>>>>>>>>> Don't think so. I have a 1 GHz scope across the laser, and didn't >>>>>>>>>>>>> see anything like that, and have applied thermal-digital >>>>>>>>>>>>> analysis(*) to the whole thing. I'll try going to 49.9 ohms and 47 >>>>>>>>>>>>> nF and see if it gets better, which it should. I might need to use >>>>>>>>>>>>> a TCA0372 booster to drive 5-ohm resistors. :( >>>>>>>>>>>>> >>>>>>>>>>>>> Cheers >>>>>>>>>>>>> >>>>>>>>>>>>> Phil Hobbs >>>>>>>>>>>>> >>>>>>>>>>>>> (*) i.e. heat gun, cold spray, and fingers >>>>>>>>>>>> >>>>>>>>>>>> Yeah, the base current noise into the 1K resistor string could make >>>>>>>>>>>> a lot of noise... assuming the signal from the opamp is perfect. >>>>>>>>>>>> >>>>>>>>>>>> Is the overall feedback optical? >>>>>>>>>>>> >>>>>>>>>>>> >>>>>>>>>>> >>>>>>>>>>> Eventually, but at the moment it's coming from the emitter resistor, so >>>>>>>>>>> that the loop doesn't null out the base current noise even within its >>>>>>>>>>> bandwidth. >>>>>>>>>>> >>>>>>>>>>> The idea is to make the bias current way sub-Poissonian, which if done >>>>>>>>>>> right results in useful amounts of amplitude squeezing of the light, >>>>>>>>>>> i.e. your detected photocurrent can actually be a few dB quieter than >>>>>>>>>>> full shot noise if you keep the optical loss low enough. It's more of a >>>>>>>>>>> checklist feature for most uses, of course, but as you've pointed out >>>>>>>>>>> there are a lot of crappy diode laser controllers out there, selling for >>>>>>>>>>> cheap, so building one that just makes the problems magically go away is >>>>>>>>>>> attractive. >>>>>>>>>>> >>>>>>>>>>> Building a new-to-me kind of high dynamic range system always seems to >>>>>>>>>>> produce a bit of buried treasure at first. The ZXTP25020 has super-low >>>>>>>>>>> Rbb' and Ree', so it would be an excellent candidate except for the 1/f >>>>>>>>>>> noise. If the higher-beta device and lower drive impedance doesn't >>>>>>>>>>> improve it enough, I'll have to try Darlingtonizing it with one of those >>>>>>>>>>> nice SiGe:C devices to return the base current to the collector circuit. >>>>>>>>>>> >>>>>>>>>>> That'd need a board spin though, and I'm super busy at the moment. >>>>>>>>>>> >>>>>>>>>>> Cheers >>>>>>>>>>> >>>>>>>>>>> Phil Hobbs >>>>>>>>>> >>>>>>>>>> How about a Sziklai pair sort of thing, with a BF862 and a PNP? Or a >>>>>>>>>> Pfet-PNP darlington? Disappear the base current. >>>>>>>>> >>>>>>>>> I need it to be a PNP-type when it's done, because the lasers are >>>>>>>>> usually grounded-cathode. (Keeping the high frequency output impedance >>>>>>>>> high is the name of the squeezing game--with an emitter follower you get >>>>>>>>> twice the shot noise.) If I could get good PFETs, that would be great, >>>>>>>>> but they're scarce these days and their noise tends to be horrible. >>>>>>>>> >>>>>>>>> The Darlington-bandaid approach could be done with an NPN and a couple >>>>>>>>> of auxiliary current sources, but it would be a bit of a mess. The >>>>>>>>> advantage would be that (done right) it would have the noise of the >>>>>>>>> BFP640, which is very low. >>>>>>>>> >>>>>>>>> Tomorrow I'll try reducing the base impedance by 10x and see if that >>>>>>>>> fixes it. Because of the 39-ohm emitter resistor, it doesn't make sense >>>>>>>>> to go very much lower than that. >>>>>>>>> >>>>>>>>> One nice thing about 1/f noise is that a 10x amplitude improvement gets >>>>>>>>> you 100x lower corner frequency. At that point I can put in a >>>>>>>>> low-frequency bandaid if I need to, such as sensing the voltage across >>>>>>>>> the base resistor and dumping the equivalent into the emitter so that >>>>>>>>> the op amp tracks out the LF noise. The trick will be to avoid a noise >>>>>>>>> peak at the crossover frequency of the bandaid. >>>>>>>>> >>>>>>>>> Cheers >>>>>>>>> >>>>>>>>> Phil Hobbs >>>>>>>> >>>>>>>> Here's an idea that I've never actually built: >>>>>>>> >>>>>>>> https://www.dropbox.com/s/lr3aekzziecz7pz/Isrc_cascode_Ib.JPG?raw=1 >>>>>>>> >>>>>>>> I don't remember what those two caps are for. >>>>>>> >>>>>>> >>>>>>> Interesting. It has the current noise of the FET at low frequency, >>>>>>> though, and looks like it has full shot noise at high frequency on >>>>>>> account of the caps. >>>>>>> >>>>>>> I'd be looking at something vaguely of that kind for the bandaid >>>>>>> approach--next time I have a Saturday afternoon free I'll have a whack >>>>>>> at it. >>>>>>> >>>>>>> Cheers >>>>>>> >>>>>>> Phil Hobbs >>>>>> >>>>>> Have you considered a bootstrap current source? That solves a lot of >>>>>> problems. Adding an inductor somewhere makes it better at high >>>>>> frequencies, so you can concentrate on the low end. Any copper TC >>>>>> error is easily compensated. >>>>> >>>>> The bootstrap sources I know about use op amps to control things at all >>>>> frequencies, so they're at least 10-15 dB noisier at medium-to-high >>>>> frequency than a transistor with Rbb' of an ohm. It's really just the >>>>> 1/f noise that's the problem. An inductor is a possibility, though--it >>>>> might very well help up around f_T. >>>>> >>>>> Cheers >>>>> >>>>> Phil Hobbs >>>> >>>> Chopamp? >>> >>> The simple op amp + BJT + local feedback approach has really stellar >>> noise performance at high frequency. The issue at low frequency is that >>> the base current 1/f noise doesn't get tracked out by the op amp loop >>> because it appears in the collector circuit but not in the emitter >>> circuit, which is where the FB is applied. >>> >>> ISTM there are four ways of fixing this wart without degrading the HF >>> noise. From easiest to hardest: >>> >>> 1. Reduce the 1/f noise by reducing the base circuit impedance. >>> >>> 2. Do some Darlingtonish thing to return the base current to the >>> collector circuit without adding significant voltage noise. >>> >>> 3. Use some more complicated local feedback scheme to synthesize a copy >>> of the base current and stick it into the emitter circuit so that the >>> main loop tracks it out. >>> >>> 4. Come up with some completely new scheme. >>> >>> Because the laser current-tunes, its linewidth is degraded more by >>> strong low-frequency components than by the same noise power spread out >>> over the full bandwidth. (This is just like an FM transmitter.) Thus >>> it would be worth trading off a bit of HF noise to fix this issue, but >>> I'm greedy and want to have it all. ;) >>> >>> Chopamps tend to have low frequency noise around 20-30 nV in 1 Hz all >>> the way down to DC, which isn't bad at all--it would flatten out my >>> noise below about 20 Hz. My faves are the OPA2188, which is a 32V dual, >>> and the OPA378, which is a bit noisier in the flatband but whose noise >>> doesn't rise at all at low frequency, at least down to 10 uHz (~1 >>> cycle/day) which is as far as I've measured it. >>> >>> I'd be very interested to hear about yours and others' fave >>> chopamps--there's a depressing amount of specsmanship going on in their >>> datasheets. >>> >>> Cheers >>> >>> Phil Hobbs >> >> We've use a bunch of them, but they tend to have high wideband noise >> and sometimes weird charge injection problems. The well-behaved >> LTC2057 is "only" 11 nv/rthz, but the ADI parts are around twice that, >> except the ADA4638, which is 66! >> >> The 2057 has a noise spike around 50 KHz, so for extreme performance >> I'd think about compounding it with another amp. Let the chopamp take >> over below the 1/f corner of another amp. > >I've done that before. One issue is that the chopamp spikes are mostly >current noise, so you have to bypass the chopamp inputs to keep the >spikes out of the rest of the circuit. >Right. The input bias current specs are all DC! I've seen weird voltage offsets, beyond the spec limits, that depend on the capacitances that the inputs see. CMOS analog switches have very interesting dynamics. I'm sort of grateful that your problem is not just now my problem! -- John Larkin Highland Technology, Inc lunatic fringe electronics
Reply by ●February 12, 20182018-02-12
On 02/12/2018 11:40 AM, John Larkin wrote:> On Mon, 12 Feb 2018 11:07:57 -0500, Phil Hobbs > <pcdhSpamMeSenseless@electrooptical.net> wrote: > >> On 02/11/2018 07:41 PM, John Larkin wrote: >>> On Sun, 11 Feb 2018 17:33:30 -0500, Phil Hobbs >>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>> >>>> On 02/11/2018 04:13 PM, John Larkin wrote: >>>>> On Sun, 11 Feb 2018 15:44:14 -0500, Phil Hobbs >>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>>>> >>>>>> On 02/11/2018 02:58 PM, John Larkin wrote: >>>>>>> On Sun, 11 Feb 2018 13:17:25 -0500, Phil Hobbs >>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>>>>>> >>>>>>>> On 02/11/2018 12:52 PM, John Larkin wrote: >>>>>>>>> On Sun, 11 Feb 2018 12:32:55 -0500, Phil Hobbs >>>>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>>>>>>>> >>>>>>>>>> On 02/11/2018 11:34 AM, John Larkin wrote: >>>>>>>>>>> On Sun, 11 Feb 2018 08:05:11 -0500, Phil Hobbs >>>>>>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>>>>>>>>>> >>>>>>>>>>>> On 02/10/2018 08:36 PM, John Larkin wrote: >>>>>>>>>>>>> On Sat, 10 Feb 2018 19:39:32 -0500, Phil Hobbs >>>>>>>>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>>>>>>>>>>>> >>>>>>>>>>>>>> On 02/10/18 18:41, John Larkin wrote: >>>>>>>>>>>>>>> On Sat, 10 Feb 2018 17:18:55 -0500, Phil Hobbs >>>>>>>>>>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: >>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> Is horrible. >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> I'm just debugging a nice diode laser controller for one >>>>>>>>>>>>>>>> customer and getting ready to do another one for another (much >>>>>>>>>>>>>>>> better) customer. It uses a ZXTP25020 PNP with a biggish >>>>>>>>>>>>>>>> degeneration resistor and two-pole bypassing of the base to get >>>>>>>>>>>>>>>> low noise at high frequency. >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> Turns out the 1/f noise is atrocious. The total noise from 2 >>>>>>>>>>>>>>>> Hz on up is only a couple of PPM, but it should be a factor of >>>>>>>>>>>>>>>> 10 better than that. With a total base resistance of 1k, the >>>>>>>>>>>>>>>> 1/f corner at 40 mA I_C is over 10 kHz. >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> The bias network looks like this: >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> +9V 0----------*--------*-------* | | | | | >>>>>>>>>>>>>>>> | | | R | 4.7 | 3.3 R 39 ohm TF | nF | >>>>>>>>>>>>>>>> nF R CCC CCC | CCC CCC | | | >>>>>>>>>>>>>>>> | / | | |< ZXTP25020 From 0--RRRR--*--RRRR--*---- >>>>>>>>>>>>>>>> | Op Amp |\ 499 499 | \ | >>>>>>>>>>>>>>>> | --- \ / --> V --> ----- | | GND >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> The emitter resistor can drop up to about 5V. >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> I've got some higher-beta transistors on order (FZT788B), and >>>>>>>>>>>>>>>> will reduce the impedance of the bias network, but I was fairly >>>>>>>>>>>>>>>> shocked that the 1/f noise was so bad. >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> Cheers >>>>>>>>>>>>>>>> >>>>>>>>>>>>>>>> Phil Hobbs >>>>>>>>>>>>>>> >>>>>>>>>>>>>>> That's a pretty fast transistor; could it be oscillating? >>>>>>>>>>>>>> >>>>>>>>>>>>>> Don't think so. I have a 1 GHz scope across the laser, and didn't >>>>>>>>>>>>>> see anything like that, and have applied thermal-digital >>>>>>>>>>>>>> analysis(*) to the whole thing. I'll try going to 49.9 ohms and 47 >>>>>>>>>>>>>> nF and see if it gets better, which it should. I might need to use >>>>>>>>>>>>>> a TCA0372 booster to drive 5-ohm resistors. :( >>>>>>>>>>>>>> >>>>>>>>>>>>>> Cheers >>>>>>>>>>>>>> >>>>>>>>>>>>>> Phil Hobbs >>>>>>>>>>>>>> >>>>>>>>>>>>>> (*) i.e. heat gun, cold spray, and fingers >>>>>>>>>>>>> >>>>>>>>>>>>> Yeah, the base current noise into the 1K resistor string could make >>>>>>>>>>>>> a lot of noise... assuming the signal from the opamp is perfect. >>>>>>>>>>>>> >>>>>>>>>>>>> Is the overall feedback optical? >>>>>>>>>>>>> >>>>>>>>>>>>> >>>>>>>>>>>> >>>>>>>>>>>> Eventually, but at the moment it's coming from the emitter resistor, so >>>>>>>>>>>> that the loop doesn't null out the base current noise even within its >>>>>>>>>>>> bandwidth. >>>>>>>>>>>> >>>>>>>>>>>> The idea is to make the bias current way sub-Poissonian, which if done >>>>>>>>>>>> right results in useful amounts of amplitude squeezing of the light, >>>>>>>>>>>> i.e. your detected photocurrent can actually be a few dB quieter than >>>>>>>>>>>> full shot noise if you keep the optical loss low enough. It's more of a >>>>>>>>>>>> checklist feature for most uses, of course, but as you've pointed out >>>>>>>>>>>> there are a lot of crappy diode laser controllers out there, selling for >>>>>>>>>>>> cheap, so building one that just makes the problems magically go away is >>>>>>>>>>>> attractive. >>>>>>>>>>>> >>>>>>>>>>>> Building a new-to-me kind of high dynamic range system always seems to >>>>>>>>>>>> produce a bit of buried treasure at first. The ZXTP25020 has super-low >>>>>>>>>>>> Rbb' and Ree', so it would be an excellent candidate except for the 1/f >>>>>>>>>>>> noise. If the higher-beta device and lower drive impedance doesn't >>>>>>>>>>>> improve it enough, I'll have to try Darlingtonizing it with one of those >>>>>>>>>>>> nice SiGe:C devices to return the base current to the collector circuit. >>>>>>>>>>>> >>>>>>>>>>>> That'd need a board spin though, and I'm super busy at the moment. >>>>>>>>>>>> >>>>>>>>>>>> Cheers >>>>>>>>>>>> >>>>>>>>>>>> Phil Hobbs >>>>>>>>>>> >>>>>>>>>>> How about a Sziklai pair sort of thing, with a BF862 and a PNP? Or a >>>>>>>>>>> Pfet-PNP darlington? Disappear the base current. >>>>>>>>>> >>>>>>>>>> I need it to be a PNP-type when it's done, because the lasers are >>>>>>>>>> usually grounded-cathode. (Keeping the high frequency output impedance >>>>>>>>>> high is the name of the squeezing game--with an emitter follower you get >>>>>>>>>> twice the shot noise.) If I could get good PFETs, that would be great, >>>>>>>>>> but they're scarce these days and their noise tends to be horrible. >>>>>>>>>> >>>>>>>>>> The Darlington-bandaid approach could be done with an NPN and a couple >>>>>>>>>> of auxiliary current sources, but it would be a bit of a mess. The >>>>>>>>>> advantage would be that (done right) it would have the noise of the >>>>>>>>>> BFP640, which is very low. >>>>>>>>>> >>>>>>>>>> Tomorrow I'll try reducing the base impedance by 10x and see if that >>>>>>>>>> fixes it. Because of the 39-ohm emitter resistor, it doesn't make sense >>>>>>>>>> to go very much lower than that. >>>>>>>>>> >>>>>>>>>> One nice thing about 1/f noise is that a 10x amplitude improvement gets >>>>>>>>>> you 100x lower corner frequency. At that point I can put in a >>>>>>>>>> low-frequency bandaid if I need to, such as sensing the voltage across >>>>>>>>>> the base resistor and dumping the equivalent into the emitter so that >>>>>>>>>> the op amp tracks out the LF noise. The trick will be to avoid a noise >>>>>>>>>> peak at the crossover frequency of the bandaid. >>>>>>>>>> >>>>>>>>>> Cheers >>>>>>>>>> >>>>>>>>>> Phil Hobbs >>>>>>>>> >>>>>>>>> Here's an idea that I've never actually built: >>>>>>>>> >>>>>>>>> https://www.dropbox.com/s/lr3aekzziecz7pz/Isrc_cascode_Ib.JPG?raw=1 >>>>>>>>> >>>>>>>>> I don't remember what those two caps are for. >>>>>>>> >>>>>>>> >>>>>>>> Interesting. It has the current noise of the FET at low frequency, >>>>>>>> though, and looks like it has full shot noise at high frequency on >>>>>>>> account of the caps. >>>>>>>> >>>>>>>> I'd be looking at something vaguely of that kind for the bandaid >>>>>>>> approach--next time I have a Saturday afternoon free I'll have a whack >>>>>>>> at it. >>>>>>>> >>>>>>>> Cheers >>>>>>>> >>>>>>>> Phil Hobbs >>>>>>> >>>>>>> Have you considered a bootstrap current source? That solves a lot of >>>>>>> problems. Adding an inductor somewhere makes it better at high >>>>>>> frequencies, so you can concentrate on the low end. Any copper TC >>>>>>> error is easily compensated. >>>>>> >>>>>> The bootstrap sources I know about use op amps to control things at all >>>>>> frequencies, so they're at least 10-15 dB noisier at medium-to-high >>>>>> frequency than a transistor with Rbb' of an ohm. It's really just the >>>>>> 1/f noise that's the problem. An inductor is a possibility, though--it >>>>>> might very well help up around f_T. >>>>>> >>>>>> Cheers >>>>>> >>>>>> Phil Hobbs >>>>> >>>>> Chopamp? >>>> >>>> The simple op amp + BJT + local feedback approach has really stellar >>>> noise performance at high frequency. The issue at low frequency is that >>>> the base current 1/f noise doesn't get tracked out by the op amp loop >>>> because it appears in the collector circuit but not in the emitter >>>> circuit, which is where the FB is applied. >>>> >>>> ISTM there are four ways of fixing this wart without degrading the HF >>>> noise. From easiest to hardest: >>>> >>>> 1. Reduce the 1/f noise by reducing the base circuit impedance. >>>> >>>> 2. Do some Darlingtonish thing to return the base current to the >>>> collector circuit without adding significant voltage noise. >>>> >>>> 3. Use some more complicated local feedback scheme to synthesize a copy >>>> of the base current and stick it into the emitter circuit so that the >>>> main loop tracks it out. >>>> >>>> 4. Come up with some completely new scheme. >>>> >>>> Because the laser current-tunes, its linewidth is degraded more by >>>> strong low-frequency components than by the same noise power spread out >>>> over the full bandwidth. (This is just like an FM transmitter.) Thus >>>> it would be worth trading off a bit of HF noise to fix this issue, but >>>> I'm greedy and want to have it all. ;) >>>> >>>> Chopamps tend to have low frequency noise around 20-30 nV in 1 Hz all >>>> the way down to DC, which isn't bad at all--it would flatten out my >>>> noise below about 20 Hz. My faves are the OPA2188, which is a 32V dual, >>>> and the OPA378, which is a bit noisier in the flatband but whose noise >>>> doesn't rise at all at low frequency, at least down to 10 uHz (~1 >>>> cycle/day) which is as far as I've measured it. >>>> >>>> I'd be very interested to hear about yours and others' fave >>>> chopamps--there's a depressing amount of specsmanship going on in their >>>> datasheets. >>>> >>>> Cheers >>>> >>>> Phil Hobbs >>> >>> We've use a bunch of them, but they tend to have high wideband noise >>> and sometimes weird charge injection problems. The well-behaved >>> LTC2057 is "only" 11 nv/rthz, but the ADI parts are around twice that, >>> except the ADA4638, which is 66! >>> >>> The 2057 has a noise spike around 50 KHz, so for extreme performance >>> I'd think about compounding it with another amp. Let the chopamp take >>> over below the 1/f corner of another amp. >> >> I've done that before. One issue is that the chopamp spikes are mostly >> current noise, so you have to bypass the chopamp inputs to keep the >> spikes out of the rest of the circuit. >> > > Right. The input bias current specs are all DC! > > I've seen weird voltage offsets, beyond the spec limits, that depend > on the capacitances that the inputs see. CMOS analog switches have > very interesting dynamics. I'm sort of grateful that your problem is > not just now my problem! > >Yeah, be that way. ;) An MPS4250 tacked onto the ZXTP25020 knocked it down pretty well, so I'll probably use that plus a bypass from the big guy's base to the supply to try to recover the quiet HF performance. Should work as long as the 4250 isn't running enough current to actually oscillate, but of course it'll dump the full shot noise of the 4250's I_C into the diode laser. That shouldn't be a gigantic issue. Hopefully the noise will be okay for Customer #1, because Customer #2's board is quite different and hasn't been fabbed yet. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC / Hobbs ElectroOptics Optics, Electro-optics, Photonics, Analog Electronics Briarcliff Manor NY 10510 http://electrooptical.net https://hobbs-eo.com
Reply by ●February 12, 20182018-02-12
On Monday, February 12, 2018 at 11:16:01 AM UTC-5, Phil Hobbs wrote:> On 02/12/2018 09:09 AM, George Herold wrote: > > On Sunday, February 11, 2018 at 5:33:47 PM UTC-5, Phil Hobbs wrote: > >> On 02/11/2018 04:13 PM, John Larkin wrote: > >>> On Sun, 11 Feb 2018 15:44:14 -0500, Phil Hobbs > >>> <pcdhSpamMeSenseless@electrooptical.net> wrote: > >>> > >>>> On 02/11/2018 02:58 PM, John Larkin wrote: > >>>>> On Sun, 11 Feb 2018 13:17:25 -0500, Phil Hobbs > >>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: > >>>>> > >>>>>> On 02/11/2018 12:52 PM, John Larkin wrote: > >>>>>>> On Sun, 11 Feb 2018 12:32:55 -0500, Phil Hobbs > >>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: > >>>>>>> > >>>>>>>> On 02/11/2018 11:34 AM, John Larkin wrote: > >>>>>>>>> On Sun, 11 Feb 2018 08:05:11 -0500, Phil Hobbs > >>>>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: > >>>>>>>>> > >>>>>>>>>> On 02/10/2018 08:36 PM, John Larkin wrote: > >>>>>>>>>>> On Sat, 10 Feb 2018 19:39:32 -0500, Phil Hobbs > >>>>>>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: > >>>>>>>>>>> > >>>>>>>>>>>> On 02/10/18 18:41, John Larkin wrote: > >>>>>>>>>>>>> On Sat, 10 Feb 2018 17:18:55 -0500, Phil Hobbs > >>>>>>>>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: > >>>>>>>>>>>>> > >>>>>>>>>>>>>> Is horrible. > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> I'm just debugging a nice diode laser controller for one > >>>>>>>>>>>>>> customer and getting ready to do another one for another (much > >>>>>>>>>>>>>> better) customer. It uses a ZXTP25020 PNP with a biggish > >>>>>>>>>>>>>> degeneration resistor and two-pole bypassing of the base to get > >>>>>>>>>>>>>> low noise at high frequency. > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> Turns out the 1/f noise is atrocious. The total noise from 2 > >>>>>>>>>>>>>> Hz on up is only a couple of PPM, but it should be a factor of > >>>>>>>>>>>>>> 10 better than that. With a total base resistance of 1k, the > >>>>>>>>>>>>>> 1/f corner at 40 mA I_C is over 10 kHz. > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> The bias network looks like this: > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> +9V 0----------*--------*-------* | | | | | > >>>>>>>>>>>>>> | | | R | 4.7 | 3.3 R 39 ohm TF | nF | > >>>>>>>>>>>>>> nF R CCC CCC | CCC CCC | | | > >>>>>>>>>>>>>> | / | | |< ZXTP25020 From 0--RRRR--*--RRRR--*---- > >>>>>>>>>>>>>> | Op Amp |\ 499 499 | \ | > >>>>>>>>>>>>>> | --- \ / --> V --> ----- | | GND > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> The emitter resistor can drop up to about 5V. > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> I've got some higher-beta transistors on order (FZT788B), and > >>>>>>>>>>>>>> will reduce the impedance of the bias network, but I was fairly > >>>>>>>>>>>>>> shocked that the 1/f noise was so bad. > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> Cheers > >>>>>>>>>>>>>> > >>>>>>>>>>>>>> Phil Hobbs > >>>>>>>>>>>>> > >>>>>>>>>>>>> That's a pretty fast transistor; could it be oscillating? > >>>>>>>>>>>> > >>>>>>>>>>>> Don't think so. I have a 1 GHz scope across the laser, and didn't > >>>>>>>>>>>> see anything like that, and have applied thermal-digital > >>>>>>>>>>>> analysis(*) to the whole thing. I'll try going to 49.9 ohms and 47 > >>>>>>>>>>>> nF and see if it gets better, which it should. I might need to use > >>>>>>>>>>>> a TCA0372 booster to drive 5-ohm resistors. :( > >>>>>>>>>>>> > >>>>>>>>>>>> Cheers > >>>>>>>>>>>> > >>>>>>>>>>>> Phil Hobbs > >>>>>>>>>>>> > >>>>>>>>>>>> (*) i.e. heat gun, cold spray, and fingers > >>>>>>>>>>> > >>>>>>>>>>> Yeah, the base current noise into the 1K resistor string could make > >>>>>>>>>>> a lot of noise... assuming the signal from the opamp is perfect. > >>>>>>>>>>> > >>>>>>>>>>> Is the overall feedback optical? > >>>>>>>>>>> > >>>>>>>>>>> > >>>>>>>>>> > >>>>>>>>>> Eventually, but at the moment it's coming from the emitter resistor, so > >>>>>>>>>> that the loop doesn't null out the base current noise even within its > >>>>>>>>>> bandwidth. > >>>>>>>>>> > >>>>>>>>>> The idea is to make the bias current way sub-Poissonian, which if done > >>>>>>>>>> right results in useful amounts of amplitude squeezing of the light, > >>>>>>>>>> i.e. your detected photocurrent can actually be a few dB quieter than > >>>>>>>>>> full shot noise if you keep the optical loss low enough. It's more of a > >>>>>>>>>> checklist feature for most uses, of course, but as you've pointed out > >>>>>>>>>> there are a lot of crappy diode laser controllers out there, selling for > >>>>>>>>>> cheap, so building one that just makes the problems magically go away is > >>>>>>>>>> attractive. > >>>>>>>>>> > >>>>>>>>>> Building a new-to-me kind of high dynamic range system always seems to > >>>>>>>>>> produce a bit of buried treasure at first. The ZXTP25020 has super-low > >>>>>>>>>> Rbb' and Ree', so it would be an excellent candidate except for the 1/f > >>>>>>>>>> noise. If the higher-beta device and lower drive impedance doesn't > >>>>>>>>>> improve it enough, I'll have to try Darlingtonizing it with one of those > >>>>>>>>>> nice SiGe:C devices to return the base current to the collector circuit. > >>>>>>>>>> > >>>>>>>>>> That'd need a board spin though, and I'm super busy at the moment. > >>>>>>>>>> > >>>>>>>>>> Cheers > >>>>>>>>>> > >>>>>>>>>> Phil Hobbs > >>>>>>>>> > >>>>>>>>> How about a Sziklai pair sort of thing, with a BF862 and a PNP? Or a > >>>>>>>>> Pfet-PNP darlington? Disappear the base current. > >>>>>>>> > >>>>>>>> I need it to be a PNP-type when it's done, because the lasers are > >>>>>>>> usually grounded-cathode. (Keeping the high frequency output impedance > >>>>>>>> high is the name of the squeezing game--with an emitter follower you get > >>>>>>>> twice the shot noise.) If I could get good PFETs, that would be great, > >>>>>>>> but they're scarce these days and their noise tends to be horrible. > >>>>>>>> > >>>>>>>> The Darlington-bandaid approach could be done with an NPN and a couple > >>>>>>>> of auxiliary current sources, but it would be a bit of a mess. The > >>>>>>>> advantage would be that (done right) it would have the noise of the > >>>>>>>> BFP640, which is very low. > >>>>>>>> > >>>>>>>> Tomorrow I'll try reducing the base impedance by 10x and see if that > >>>>>>>> fixes it. Because of the 39-ohm emitter resistor, it doesn't make sense > >>>>>>>> to go very much lower than that. > >>>>>>>> > >>>>>>>> One nice thing about 1/f noise is that a 10x amplitude improvement gets > >>>>>>>> you 100x lower corner frequency. At that point I can put in a > >>>>>>>> low-frequency bandaid if I need to, such as sensing the voltage across > >>>>>>>> the base resistor and dumping the equivalent into the emitter so that > >>>>>>>> the op amp tracks out the LF noise. The trick will be to avoid a noise > >>>>>>>> peak at the crossover frequency of the bandaid. > >>>>>>>> > >>>>>>>> Cheers > >>>>>>>> > >>>>>>>> Phil Hobbs > >>>>>>> > >>>>>>> Here's an idea that I've never actually built: > >>>>>>> > >>>>>>> https://www.dropbox.com/s/lr3aekzziecz7pz/Isrc_cascode_Ib.JPG?raw=1 > >>>>>>> > >>>>>>> I don't remember what those two caps are for. > >>>>>> > >>>>>> > >>>>>> Interesting. It has the current noise of the FET at low frequency, > >>>>>> though, and looks like it has full shot noise at high frequency on > >>>>>> account of the caps. > >>>>>> > >>>>>> I'd be looking at something vaguely of that kind for the bandaid > >>>>>> approach--next time I have a Saturday afternoon free I'll have a whack > >>>>>> at it. > >>>>>> > >>>>>> Cheers > >>>>>> > >>>>>> Phil Hobbs > >>>>> > >>>>> Have you considered a bootstrap current source? That solves a lot of > >>>>> problems. Adding an inductor somewhere makes it better at high > >>>>> frequencies, so you can concentrate on the low end. Any copper TC > >>>>> error is easily compensated. > >>>> > >>>> The bootstrap sources I know about use op amps to control things at all > >>>> frequencies, so they're at least 10-15 dB noisier at medium-to-high > >>>> frequency than a transistor with Rbb' of an ohm. It's really just the > >>>> 1/f noise that's the problem. An inductor is a possibility, though--it > >>>> might very well help up around f_T. > >>>> > >>>> Cheers > >>>> > >>>> Phil Hobbs > >>> > >>> Chopamp? > >> > > > >> The simple op amp + BJT + local feedback approach has really stellar > >> noise performance at high frequency. The issue at low frequency is that > >> the base current 1/f noise doesn't get tracked out by the op amp loop > >> because it appears in the collector circuit but not in the emitter > >> circuit, which is where the FB is applied. > > > > Phil, the above bothered me. Isn't the emitter current the sum of > > base and collector? But I think I understand... base current 1/f > > noise appears in emitter, and gets taken care of by FB... but that > > screws up the collector current which has no 1/f noise. > > Or is my head on backwards this morning? > > You can look at it either way. Inside the op amp loop's control > bandwidth, the voltage across the emitter resistor is held still by > overall feedback, so regardless of the noise voltage drop across the > base resistor, all the 1/f noise has to come out the collector.Thanks, I thought that, but it's good to check. I use the (pfet) VP2106, I < = 100mA. I never looked at the noise closely. the loose spec we quote is 50 nA (in 100 kHz?) which is .. ~20X the shot noise. George H.> > Outside that bandwidth, there's a narrow region where the base resistors > contribute, but then the bypass caps take over and hold the voltage > across the emitter resistance constant again. So basically the noise > comes out the collector at all frequencies as long as the emitter > resistor is dropping several times kT/e. > > > >> > >> ISTM there are four ways of fixing this wart without degrading the HF > >> noise. From easiest to hardest: > >> > >> 1. Reduce the 1/f noise by reducing the base circuit impedance. > > Didn't help, for reasons given above. > > >> > >> 2. Do some Darlingtonish thing to return the base current to the > >> collector circuit without adding significant voltage noise. > > Using an MPS4250 Darlington helped a lot with the 1/f noise, but the > flatband noise is now only about 6 dB below shot noise, which is a pity. > Good enough for this rev, though. Now I just need to find a suitable > Darlington in SOT89 or SOT223. > > Next time round I'll use another of the power devices as the input > transistor, and add some extra base current to it, but not enough for > its 1/f noise to dominate. > > > >> > >> 3. Use some more complicated local feedback scheme to synthesize a copy > >> of the base current and stick it into the emitter circuit so that the > >> main loop tracks it out. > >> > >> 4. Come up with some completely new scheme. > > > > You could feed back with a sense resistor on the same side as the load. > > I guess that means either floating the load (laser diode.. not good) > > or floating the sense R. (or I'm confused) > > > > George H. > > -- > Dr Philip C D Hobbs > Principal Consultant > ElectroOptical Innovations LLC / Hobbs ElectroOptics > Optics, Electro-optics, Photonics, Analog Electronics > Briarcliff Manor NY 10510 > > http://electrooptical.net > https://hobbs-eo.com
Reply by ●February 12, 20182018-02-12
On Monday, February 12, 2018 at 12:03:49 PM UTC-5, George Herold wrote:> On Monday, February 12, 2018 at 11:16:01 AM UTC-5, Phil Hobbs wrote: > > On 02/12/2018 09:09 AM, George Herold wrote: > > > On Sunday, February 11, 2018 at 5:33:47 PM UTC-5, Phil Hobbs wrote: > > >> On 02/11/2018 04:13 PM, John Larkin wrote: > > >>> On Sun, 11 Feb 2018 15:44:14 -0500, Phil Hobbs > > >>> <pcdhSpamMeSenseless@electrooptical.net> wrote: > > >>> > > >>>> On 02/11/2018 02:58 PM, John Larkin wrote: > > >>>>> On Sun, 11 Feb 2018 13:17:25 -0500, Phil Hobbs > > >>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: > > >>>>> > > >>>>>> On 02/11/2018 12:52 PM, John Larkin wrote: > > >>>>>>> On Sun, 11 Feb 2018 12:32:55 -0500, Phil Hobbs > > >>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: > > >>>>>>> > > >>>>>>>> On 02/11/2018 11:34 AM, John Larkin wrote: > > >>>>>>>>> On Sun, 11 Feb 2018 08:05:11 -0500, Phil Hobbs > > >>>>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: > > >>>>>>>>> > > >>>>>>>>>> On 02/10/2018 08:36 PM, John Larkin wrote: > > >>>>>>>>>>> On Sat, 10 Feb 2018 19:39:32 -0500, Phil Hobbs > > >>>>>>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: > > >>>>>>>>>>> > > >>>>>>>>>>>> On 02/10/18 18:41, John Larkin wrote: > > >>>>>>>>>>>>> On Sat, 10 Feb 2018 17:18:55 -0500, Phil Hobbs > > >>>>>>>>>>>>> <pcdhSpamMeSenseless@electrooptical.net> wrote: > > >>>>>>>>>>>>> > > >>>>>>>>>>>>>> Is horrible. > > >>>>>>>>>>>>>> > > >>>>>>>>>>>>>> I'm just debugging a nice diode laser controller for one > > >>>>>>>>>>>>>> customer and getting ready to do another one for another (much > > >>>>>>>>>>>>>> better) customer. It uses a ZXTP25020 PNP with a biggish > > >>>>>>>>>>>>>> degeneration resistor and two-pole bypassing of the base to get > > >>>>>>>>>>>>>> low noise at high frequency. > > >>>>>>>>>>>>>> > > >>>>>>>>>>>>>> Turns out the 1/f noise is atrocious. The total noise from 2 > > >>>>>>>>>>>>>> Hz on up is only a couple of PPM, but it should be a factor of > > >>>>>>>>>>>>>> 10 better than that. With a total base resistance of 1k, the > > >>>>>>>>>>>>>> 1/f corner at 40 mA I_C is over 10 kHz. > > >>>>>>>>>>>>>> > > >>>>>>>>>>>>>> The bias network looks like this: > > >>>>>>>>>>>>>> > > >>>>>>>>>>>>>> +9V 0----------*--------*-------* | | | | | > > >>>>>>>>>>>>>> | | | R | 4.7 | 3.3 R 39 ohm TF | nF | > > >>>>>>>>>>>>>> nF R CCC CCC | CCC CCC | | | > > >>>>>>>>>>>>>> | / | | |< ZXTP25020 From 0--RRRR--*--RRRR--*---- > > >>>>>>>>>>>>>> | Op Amp |\ 499 499 | \ | > > >>>>>>>>>>>>>> | --- \ / --> V --> ----- | | GND > > >>>>>>>>>>>>>> > > >>>>>>>>>>>>>> The emitter resistor can drop up to about 5V. > > >>>>>>>>>>>>>> > > >>>>>>>>>>>>>> I've got some higher-beta transistors on order (FZT788B), and > > >>>>>>>>>>>>>> will reduce the impedance of the bias network, but I was fairly > > >>>>>>>>>>>>>> shocked that the 1/f noise was so bad. > > >>>>>>>>>>>>>> > > >>>>>>>>>>>>>> Cheers > > >>>>>>>>>>>>>> > > >>>>>>>>>>>>>> Phil Hobbs > > >>>>>>>>>>>>> > > >>>>>>>>>>>>> That's a pretty fast transistor; could it be oscillating? > > >>>>>>>>>>>> > > >>>>>>>>>>>> Don't think so. I have a 1 GHz scope across the laser, and didn't > > >>>>>>>>>>>> see anything like that, and have applied thermal-digital > > >>>>>>>>>>>> analysis(*) to the whole thing. I'll try going to 49.9 ohms and 47 > > >>>>>>>>>>>> nF and see if it gets better, which it should. I might need to use > > >>>>>>>>>>>> a TCA0372 booster to drive 5-ohm resistors. :( > > >>>>>>>>>>>> > > >>>>>>>>>>>> Cheers > > >>>>>>>>>>>> > > >>>>>>>>>>>> Phil Hobbs > > >>>>>>>>>>>> > > >>>>>>>>>>>> (*) i.e. heat gun, cold spray, and fingers > > >>>>>>>>>>> > > >>>>>>>>>>> Yeah, the base current noise into the 1K resistor string could make > > >>>>>>>>>>> a lot of noise... assuming the signal from the opamp is perfect. > > >>>>>>>>>>> > > >>>>>>>>>>> Is the overall feedback optical? > > >>>>>>>>>>> > > >>>>>>>>>>> > > >>>>>>>>>> > > >>>>>>>>>> Eventually, but at the moment it's coming from the emitter resistor, so > > >>>>>>>>>> that the loop doesn't null out the base current noise even within its > > >>>>>>>>>> bandwidth. > > >>>>>>>>>> > > >>>>>>>>>> The idea is to make the bias current way sub-Poissonian, which if done > > >>>>>>>>>> right results in useful amounts of amplitude squeezing of the light, > > >>>>>>>>>> i.e. your detected photocurrent can actually be a few dB quieter than > > >>>>>>>>>> full shot noise if you keep the optical loss low enough. It's more of a > > >>>>>>>>>> checklist feature for most uses, of course, but as you've pointed out > > >>>>>>>>>> there are a lot of crappy diode laser controllers out there, selling for > > >>>>>>>>>> cheap, so building one that just makes the problems magically go away is > > >>>>>>>>>> attractive. > > >>>>>>>>>> > > >>>>>>>>>> Building a new-to-me kind of high dynamic range system always seems to > > >>>>>>>>>> produce a bit of buried treasure at first. The ZXTP25020 has super-low > > >>>>>>>>>> Rbb' and Ree', so it would be an excellent candidate except for the 1/f > > >>>>>>>>>> noise. If the higher-beta device and lower drive impedance doesn't > > >>>>>>>>>> improve it enough, I'll have to try Darlingtonizing it with one of those > > >>>>>>>>>> nice SiGe:C devices to return the base current to the collector circuit. > > >>>>>>>>>> > > >>>>>>>>>> That'd need a board spin though, and I'm super busy at the moment. > > >>>>>>>>>> > > >>>>>>>>>> Cheers > > >>>>>>>>>> > > >>>>>>>>>> Phil Hobbs > > >>>>>>>>> > > >>>>>>>>> How about a Sziklai pair sort of thing, with a BF862 and a PNP? Or a > > >>>>>>>>> Pfet-PNP darlington? Disappear the base current. > > >>>>>>>> > > >>>>>>>> I need it to be a PNP-type when it's done, because the lasers are > > >>>>>>>> usually grounded-cathode. (Keeping the high frequency output impedance > > >>>>>>>> high is the name of the squeezing game--with an emitter follower you get > > >>>>>>>> twice the shot noise.) If I could get good PFETs, that would be great, > > >>>>>>>> but they're scarce these days and their noise tends to be horrible. > > >>>>>>>> > > >>>>>>>> The Darlington-bandaid approach could be done with an NPN and a couple > > >>>>>>>> of auxiliary current sources, but it would be a bit of a mess. The > > >>>>>>>> advantage would be that (done right) it would have the noise of the > > >>>>>>>> BFP640, which is very low. > > >>>>>>>> > > >>>>>>>> Tomorrow I'll try reducing the base impedance by 10x and see if that > > >>>>>>>> fixes it. Because of the 39-ohm emitter resistor, it doesn't make sense > > >>>>>>>> to go very much lower than that. > > >>>>>>>> > > >>>>>>>> One nice thing about 1/f noise is that a 10x amplitude improvement gets > > >>>>>>>> you 100x lower corner frequency. At that point I can put in a > > >>>>>>>> low-frequency bandaid if I need to, such as sensing the voltage across > > >>>>>>>> the base resistor and dumping the equivalent into the emitter so that > > >>>>>>>> the op amp tracks out the LF noise. The trick will be to avoid a noise > > >>>>>>>> peak at the crossover frequency of the bandaid. > > >>>>>>>> > > >>>>>>>> Cheers > > >>>>>>>> > > >>>>>>>> Phil Hobbs > > >>>>>>> > > >>>>>>> Here's an idea that I've never actually built: > > >>>>>>> > > >>>>>>> https://www.dropbox.com/s/lr3aekzziecz7pz/Isrc_cascode_Ib.JPG?raw=1 > > >>>>>>> > > >>>>>>> I don't remember what those two caps are for. > > >>>>>> > > >>>>>> > > >>>>>> Interesting. It has the current noise of the FET at low frequency, > > >>>>>> though, and looks like it has full shot noise at high frequency on > > >>>>>> account of the caps. > > >>>>>> > > >>>>>> I'd be looking at something vaguely of that kind for the bandaid > > >>>>>> approach--next time I have a Saturday afternoon free I'll have a whack > > >>>>>> at it. > > >>>>>> > > >>>>>> Cheers > > >>>>>> > > >>>>>> Phil Hobbs > > >>>>> > > >>>>> Have you considered a bootstrap current source? That solves a lot of > > >>>>> problems. Adding an inductor somewhere makes it better at high > > >>>>> frequencies, so you can concentrate on the low end. Any copper TC > > >>>>> error is easily compensated. > > >>>> > > >>>> The bootstrap sources I know about use op amps to control things at all > > >>>> frequencies, so they're at least 10-15 dB noisier at medium-to-high > > >>>> frequency than a transistor with Rbb' of an ohm. It's really just the > > >>>> 1/f noise that's the problem. An inductor is a possibility, though--it > > >>>> might very well help up around f_T. > > >>>> > > >>>> Cheers > > >>>> > > >>>> Phil Hobbs > > >>> > > >>> Chopamp? > > >> > > > > > >> The simple op amp + BJT + local feedback approach has really stellar > > >> noise performance at high frequency. The issue at low frequency is that > > >> the base current 1/f noise doesn't get tracked out by the op amp loop > > >> because it appears in the collector circuit but not in the emitter > > >> circuit, which is where the FB is applied. > > > > > > Phil, the above bothered me. Isn't the emitter current the sum of > > > base and collector? But I think I understand... base current 1/f > > > noise appears in emitter, and gets taken care of by FB... but that > > > screws up the collector current which has no 1/f noise. > > > Or is my head on backwards this morning? > > > > You can look at it either way. Inside the op amp loop's control > > bandwidth, the voltage across the emitter resistor is held still by > > overall feedback, so regardless of the noise voltage drop across the > > base resistor, all the 1/f noise has to come out the collector. > > Thanks, I thought that, but it's good to check. > I use the (pfet) VP2106, I < = 100mA. I never looked at the noise > closely. the loose spec we quote is 50 nA (in 100 kHz?) which is > .. ~20X the shot noise.Driving home.. you know I did measure the noise once. I had two lasers running. I locked one to the edge of an atomic absorption line, (sharp edge). Tuned the other to the same spot.. but left it free running. sent them both into a ~50/50 beam splitter,(mixed) and onto a fast photodiode, I'd get a nice beat note ( ~10 MHz) the 'scope FFT would give a BW of ~100 kHz.. maybe 200, mostly instrument limited. 100 kHz out of ~3.8x 10^14 Hz (~780 nm) ~is 10^-9 territory. There's some factor relating current to wavelength, I'll have to look it up. Anyway the beat note was intermittent, and would disappear with bumps and other disturbances, I wonder if 1/f current noise was part of the problem.... (This would have been in pfet's.) George H.> > George H. > > > > Outside that bandwidth, there's a narrow region where the base resistors > > contribute, but then the bypass caps take over and hold the voltage > > across the emitter resistance constant again. So basically the noise > > comes out the collector at all frequencies as long as the emitter > > resistor is dropping several times kT/e. > > > > > > >> > > >> ISTM there are four ways of fixing this wart without degrading the HF > > >> noise. From easiest to hardest: > > >> > > >> 1. Reduce the 1/f noise by reducing the base circuit impedance. > > > > Didn't help, for reasons given above. > > > > >> > > >> 2. Do some Darlingtonish thing to return the base current to the > > >> collector circuit without adding significant voltage noise. > > > > Using an MPS4250 Darlington helped a lot with the 1/f noise, but the > > flatband noise is now only about 6 dB below shot noise, which is a pity. > > Good enough for this rev, though. Now I just need to find a suitable > > Darlington in SOT89 or SOT223. > > > > Next time round I'll use another of the power devices as the input > > transistor, and add some extra base current to it, but not enough for > > its 1/f noise to dominate. > > > > > > >> > > >> 3. Use some more complicated local feedback scheme to synthesize a copy > > >> of the base current and stick it into the emitter circuit so that the > > >> main loop tracks it out. > > >> > > >> 4. Come up with some completely new scheme. > > > > > > You could feed back with a sense resistor on the same side as the load. > > > I guess that means either floating the load (laser diode.. not good) > > > or floating the sense R. (or I'm confused) > > > > > > George H. > > > > -- > > Dr Philip C D Hobbs > > Principal Consultant > > ElectroOptical Innovations LLC / Hobbs ElectroOptics > > Optics, Electro-optics, Photonics, Analog Electronics > > Briarcliff Manor NY 10510 > > > > http://electrooptical.net > > https://hobbs-eo.com
Reply by ●February 15, 20182018-02-15
On 02/10/2018 05:18 PM, Phil Hobbs wrote:> Is horrible. > > I'm just debugging a nice diode laser controller for one customer and > getting ready to do another one for another (much better) customer. It > uses a ZXTP25020 PNP with a biggish degeneration resistor and two-pole > bypassing of the base to get low noise at high frequency. > > Turns out the 1/f noise is atrocious. The total noise from 2 Hz on up > is only a couple of PPM, but it should be a factor of 10 better than > that. With a total base resistance of 1k, the 1/f corner at 40 mA I_C > is over 10 kHz. > > The bias network looks like this: > > +9V 0----------*--------*-------* > | | | > | | | > | | R > | 4.7 | 3.3 R 39 ohm TF > | nF | nF R > CCC CCC | > CCC CCC | > | | | / > | | |< ZXTP25020 > From 0--RRRR--*--RRRR--*---- | > Op Amp |\ > 499 499 | \ > | > | > --- > \ / --> > V --> > ----- > | > | > GND > > The emitter resistor can drop up to about 5V. > > I've got some higher-beta transistors on order (FZT788B), and will > reduce the impedance of the bias network, but I was fairly shocked that > the 1/f noise was so bad. > > Cheers > > Phil Hobbs ><egg-on-face> Turned out that most of the 1/f noise was due to the HP 50-ohm termination I was using as a test load--which turned out to be made of cermet--plus the temperature coefficient of the leakage of a 22-uF 50V aluminum polymer cap I was using to AC-couple the noise measurement amplifier. I found this out by replacing the laser driver with an HP precision power supply (set to 80V) and a 2k wirewound resistor. The two effects were almost the same size at the lower frequency limit of my measurement, so nothing I did changed it much, until I fixed both at once. The 1/f corner is now down around 20 Hz where it belongs, and the box is way sub-poissonian outside the op amp bandwidth. </egg-on-face> Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC / Hobbs ElectroOptics Optics, Electro-optics, Photonics, Analog Electronics Briarcliff Manor NY 10510 http://electrooptical.net https://hobbs-eo.com
Reply by ●February 15, 20182018-02-15
On Thu, 15 Feb 2018 15:00:26 -0500, Phil Hobbs <pcdhSpamMeSenseless@electrooptical.net> wrote:>On 02/10/2018 05:18 PM, Phil Hobbs wrote: >> Is horrible. >> >> I'm just debugging a nice diode laser controller for one customer and >> getting ready to do another one for another (much better) customer. It >> uses a ZXTP25020 PNP with a biggish degeneration resistor and two-pole >> bypassing of the base to get low noise at high frequency. >> >> Turns out the 1/f noise is atrocious. The total noise from 2 Hz on up >> is only a couple of PPM, but it should be a factor of 10 better than >> that. With a total base resistance of 1k, the 1/f corner at 40 mA I_C >> is over 10 kHz. >> >> The bias network looks like this: >> >> +9V 0----------*--------*-------* >> | | | >> | | | >> | | R >> | 4.7 | 3.3 R 39 ohm TF >> | nF | nF R >> CCC CCC | >> CCC CCC | >> | | | / >> | | |< ZXTP25020 >> From 0--RRRR--*--RRRR--*---- | >> Op Amp |\ >> 499 499 | \ >> | >> | >> --- >> \ / --> >> V --> >> ----- >> | >> | >> GND >> >> The emitter resistor can drop up to about 5V. >> >> I've got some higher-beta transistors on order (FZT788B), and will >> reduce the impedance of the bias network, but I was fairly shocked that >> the 1/f noise was so bad. >> >> Cheers >> >> Phil Hobbs >> > ><egg-on-face> > >Turned out that most of the 1/f noise was due to the HP 50-ohm >termination I was using as a test load--which turned out to be made of >cermet--plus the temperature coefficient of the leakage of a 22-uF 50V >aluminum polymer cap I was using to AC-couple the noise measurement >amplifier. > >I found this out by replacing the laser driver with an HP precision >power supply (set to 80V) and a 2k wirewound resistor. > >The two effects were almost the same size at the lower frequency limit >of my measurement, so nothing I did changed it much, until I fixed both >at once. > >The 1/f corner is now down around 20 Hz where it belongs, and the box is >way sub-poissonian outside the op amp bandwidth. > ></egg-on-face> > >Cheers > >Phil HobbsYou should have caught that long ago when I asked if you had run a noise vs device parameter simulation in LTspice ?>:-} ...Jim Thompson -- | James E.Thompson | mens | | Analog Innovations | et | | Analog/Mixed-Signal ASIC's and Discrete Systems | manus | | STV, Queen Creek, AZ 85142 Skype: skypeanalog | | | Voice:(480)460-2350 Fax: Available upon request | Brass Rat | | E-mail Icon at http://www.analog-innovations.com | 1962 | It's what you learn, after you know it all, that counts.
Reply by ●February 15, 20182018-02-15
On Thursday, February 15, 2018 at 3:00:33 PM UTC-5, Phil Hobbs wrote:> On 02/10/2018 05:18 PM, Phil Hobbs wrote: > > Is horrible. > > > > I'm just debugging a nice diode laser controller for one customer and > > getting ready to do another one for another (much better) customer. It > > uses a ZXTP25020 PNP with a biggish degeneration resistor and two-pole > > bypassing of the base to get low noise at high frequency. > > > > Turns out the 1/f noise is atrocious. The total noise from 2 Hz on up > > is only a couple of PPM, but it should be a factor of 10 better than > > that. With a total base resistance of 1k, the 1/f corner at 40 mA I_C > > is over 10 kHz. > > > > The bias network looks like this: > > > > +9V 0----------*--------*-------* > > | | | > > | | | > > | | R > > | 4.7 | 3.3 R 39 ohm TF > > | nF | nF R > > CCC CCC | > > CCC CCC | > > | | | / > > | | |< ZXTP25020 > > From 0--RRRR--*--RRRR--*---- | > > Op Amp |\ > > 499 499 | \ > > | > > | > > --- > > \ / --> > > V --> > > ----- > > | > > | > > GND > > > > The emitter resistor can drop up to about 5V. > > > > I've got some higher-beta transistors on order (FZT788B), and will > > reduce the impedance of the bias network, but I was fairly shocked that > > the 1/f noise was so bad. > > > > Cheers > > > > Phil Hobbs > > > > <egg-on-face> > > Turned out that most of the 1/f noise was due to the HP 50-ohm > termination I was using as a test load--which turned out to be made of > cermet--plus the temperature coefficient of the leakage of a 22-uF 50V > aluminum polymer cap I was using to AC-couple the noise measurement > amplifier. > > I found this out by replacing the laser driver with an HP precision > power supply (set to 80V) and a 2k wirewound resistor. > > The two effects were almost the same size at the lower frequency limit > of my measurement, so nothing I did changed it much, until I fixed both > at once. > > The 1/f corner is now down around 20 Hz where it belongs, and the box is > way sub-poissonian outside the op amp bandwidth. > > </egg-on-face>Eggcellent. (Thanks for sharing.) When looking at noise, I made the same mistake twice! And only found it when I went back to my notebook.. saying to myself, "I did this two years ago and it worked.." Big dope slap. So what do you use for a coupling cap? And do you guesstimate the 20 Hz corner.. (since 50 ohms and 22 uF is 1^10-3 sec, ~100 Hz.) George h.> > Cheers > > Phil Hobbs > > -- > Dr Philip C D Hobbs > Principal Consultant > ElectroOptical Innovations LLC / Hobbs ElectroOptics > Optics, Electro-optics, Photonics, Analog Electronics > Briarcliff Manor NY 10510 > > http://electrooptical.net > https://hobbs-eo.com
